Real-time Particle Filter Tracking Method Based on Human Vision Non-uniform Characteristics
CHEN Yi,SUN Xiao-wei,LI Yan-jun
DOI: https://doi.org/10.3969/j.issn.1001-8891.2010.11.001
2010-01-01
Abstract:Because plenty of information needs to be dealt with during the process of image-guiding,in order to retain effective information as much as possible and to lower the amount of calculation,a human visual model of non-uniformity sampling,Log-polar coordinate model,is used,to compress the amount of information and to improve the speed of computing.In addition,because of the invariance of the Log-polar coordinate transformation on the target shape with the rotation and scaling,the proposed model is very robust in tracking non-rigid targets.Taking the last stage in the image tracking into account,centroid and corners of the class tracking method will produce corner's drift,in order to suppress corner's drift,the Mean Shift tracking method is used,which is based on target strength characteristics.Taking the complex environment of the target tracking with noise and clutter into account,and the target model of non-Gaussian non-linear in nature,the human eyes non-uniformity sampling model is used.In our approach the Mean Shift tracking method and particle filter are combined and the algorithm is simulated.The experimental results show that the method can effectively suppress match-points drifting,reduce the calculation cost of target tracking,and is a robust target tracking method.